Chronos-2: From Univariate to Universal Forecasting
Abdul Fatir Ansari, Oleksandr Shchur, Jaris Küken, Andreas Auer, Boran Han, Pedro Mercado, Syama Sundar Rangapuram, Huibin Shen, Lorenzo Stella, Xiyuan Zhang, Mononito Goswami, Shubham Kapoor, Danielle C. Maddix, Pablo Guerron, Tony Hu, Junming Yin, Nick Erickson, Prateek Mutalik Desai, Hao Wang, Huzefa Rangwala, George Karypis, Yuyang Wang
2025-10-21
Summary
This paper introduces Chronos-2, a new forecasting model that can make predictions about future data without needing to be specifically trained for each new task. It's designed to handle different types of forecasting problems, including those with multiple related data streams and external factors that influence the predictions.
What's the problem?
Existing forecasting models are often built for very specific situations. Most pretrained models only work well when predicting a single thing over time (univariate forecasting). Real-world problems usually involve predicting multiple things at once (multivariate) and considering other information (covariates) that affects the outcome. Current models struggle to adapt to these more complex scenarios without a lot of retraining.
What's the solution?
The researchers created Chronos-2, which uses a special 'group attention' mechanism. This allows the model to learn from similar time series data together, even if they represent different but related things. They trained Chronos-2 on artificially created datasets that had complex relationships between different data streams, which helped it learn to handle real-world complexity. This means you can give Chronos-2 a new forecasting problem, and it can figure it out 'in context' without needing to be retrained.
Why it matters?
Chronos-2 is a significant step towards a general-purpose forecasting model. It performs very well on standard tests, especially when dealing with multiple variables and external factors. This means businesses in areas like energy and retail can use it directly to make better predictions without the time and expense of custom model building and training, making forecasting more accessible and efficient.
Abstract
Pretrained time series models have enabled inference-only forecasting systems that produce accurate predictions without task-specific training. However, existing approaches largely focus on univariate forecasting, limiting their applicability in real-world scenarios where multivariate data and covariates play a crucial role. We present Chronos-2, a pretrained model capable of handling univariate, multivariate, and covariate-informed forecasting tasks in a zero-shot manner. Chronos-2 employs a group attention mechanism that facilitates in-context learning (ICL) through efficient information sharing across multiple time series within a group, which may represent sets of related series, variates of a multivariate series, or targets and covariates in a forecasting task. These general capabilities are achieved through training on synthetic datasets that impose diverse multivariate structures on univariate series. Chronos-2 delivers state-of-the-art performance across three comprehensive benchmarks: fev-bench, GIFT-Eval, and Chronos Benchmark II. On fev-bench, which emphasizes multivariate and covariate-informed forecasting, Chronos-2's universal ICL capabilities lead to substantial improvements over existing models. On tasks involving covariates, it consistently outperforms baselines by a wide margin. Case studies in the energy and retail domains further highlight its practical advantages. The in-context learning capabilities of Chronos-2 establish it as a general-purpose forecasting model that can be used "as is" in real-world forecasting pipelines.